Layers

Starting with version 0.17, a new concept of Layers has been added to ImFusion Labels. Instead of having a one-to-one association between a dataset and a type of annotation, there can now be multiple sets of annotations associated to each dataset. Each layer is individually defined with its own labels, and only a single layer is active at a time.

Defining layers

Defining the layers of a project is done in the settings, by actvating the Expert feature level. Two tables are then shown on the right-hand side: the top one is used to define how many layers there are, and the bottom one is used to define the content of the currently selected layer.

_images/layer-definition.png

In the above example, three pixelwise segmentations layers are defined, the first one is selected.

The labels of each layer are completely independent.

Changing layers

If a project has been defined with multiple layers, a button to switch the active layers is shown in the database view, in the top right of the window. When clicked, it shows a drop-down with the list of all layers, which can then be clicked to change the active layers. The same button is shown at the bottom-right of the annotation view.

_images/layer-switch.png

In this example, the project only contains pixelwise segmentation. It has two layers, one named ‘Lungs’ and the other ‘Lesions’.

As the active layer is changed, the thumbnail of each dataset is updated to show the annotation corresponding to the newly selected layer. If a dataset is currently loaded, the annotation is also switched to correspond to the active layers. Note that the previously active annotation is automatically saved.

Exporting with layers

When exporting data of a project with layers, the layer to be exported must be selected in the Label Encoding section. At the moment, only one layer can be exported at a time, but this may change in the future.

Note

One notable use-case for this feature is pixelwise segmentation of overlapping features. Within one layer, only a single label can be assigned to one pixel, but with multiple layers the overlapping features can be annotated. For example, if one wanted to segment lungs and lesions in a chest CT-scan, one possibility would be to define two layers: the first representing the entire lung, and the second layer representing the lesions, which are a subset of the lungs.